VideoSAR moving target detection and tracking algorithm based on deep learning
-
摘要: 视频合成孔径雷达(synthetic aperture Radar,SAR)技术被广泛应用于军事侦查、地质勘探和灾害预测等领域。由于SAR视频存在很多的相干斑(Speckle)噪声以及镜面反射、叠掩效应等干扰因素,运动目标容易与背景或其他目标混淆在一起。针对上述问题,文章提出了一种有效的视频SAR目标检测与跟踪算法。首先,提取视频SAR的多个特征用于构造多通道特征图; 然后,使用改进的轻量EfficientDet网络对更深层的特征进行提取,从而在兼顾算法效率的同时提升SAR目标检测的准确度; 最后,采用基于目标检测框的轨迹关联策略对视频SAR中同一目标进行关联。实验表明,本研究提出的方法针对SAR阴影目标检测与跟踪任务取得了较好的效果。Abstract: The video synthetic aperture radar (VideoSAR) technology is widely used in military reconnaissance, geological exploration, and disaster prediction, among other fields. Owing to multiple interference factors in SAR videos, such as speckle noise, specular reflection, and overlay effect, moving targets are easily mixed with background or other targets. Therefore, this study proposed an effective VideoSAR target detection and tracking algorithm. Firstly, several features of VideoSAR were extracted to construct multichannel feature maps. Then, deeper features were extracted using the improved lightweight EfficientDet network, thus improving the accuracy of SAR target detection while considering algorithm efficiency. Finally, the trajectory association strategy based on bounding boxes was employed to associate the same target in VideoSAR. The experimental results show that the method proposed in this study is effective for SAR shadow target detection and tracking.
-
Key words:
- VideoSAR /
- feature enhancement /
- target detection /
- deep learning /
- feature pyramid network /
- multi-target tracking
-
-
计量
- 文章访问数: 36
- PDF下载数: 6
- 施引文献: 0
下载: